Performance Comparison of Image Segmentation Techniques for Lung Nodule Detection in CT Images

Lung cancer is the most deadliest disease all over the world. It is caused by uncontrolled growth of abnormal cells which leads to formation of lumps called nodules in the lung. Now days, the image processing techniques are extensively used in numerous medical areas to increase the survival rate. The paper includes comparison between three segmentation techniques namely iterative thresholding, Region and Fuzzy Region based level set method. A standardized LIDC dataset is used to analyze the performance of segmentation techniques with respect to different type of nodules (well circumscribed and pleura attached). Experimental investigations show that iterative thresholding method detects well circumscribed nodules with high degree of accuracy but fails to detect pleura attached nodules due to inaccurate extraction of boundary of nodules in some cases. LSM handles boundary leakage problem and perfectly detects pleura attached nodules in case of Region Based Level Set Method (RBLSM) and Fuzzy Region Based Level Set Method (FRBLSM). But RBLSM may not be able to detect well circumscribed nodules in some CT slices whereas the overall performance of FRBLSM is better in terms of False Positive (FP) and True Positive (TP).